To the Editor.—
The recent commentary by Shrout and colleagues1 brings an interesting historical perspective to those articles in the Archives aimed at quantifying agreement in psychiatric diagnosis. The k statistic was explicated 18 years ago,2 shown to have some nonobvious properties in indexing reliability and the effect of reliability on validity,3 criticized by some,4 supplanted with another statistic,5 and, phoenixlike, resurrected.1 Lest any further debate erupt over k or other statistics, I suggest that we may be entirely on the wrong track.Statistics such as k, Yule's Y, or even sensitivity and specificity are single numbers used to parsimoniously explain a large amount of data. The statistic is derived from a formal mathematical model that represents diagnostic agreement in the real world. The controversy centers around how well these statistics summarize information from the formal mathematical model under variable base rates. There is
Carey G. Y's, k'S, p's and q's. Arch Gen Psychiatry. 1987;44(11):1027. doi:10.1001/archpsyc.1987.01800230107020
Customize your JAMA Network experience by selecting one or more topics from the list below.
Create a personal account or sign in to: